基于矩形特征的无空间参考DEM图像模糊匹配

摘要

One of the key technologies of DEM application is multi‐source ,multi‐temporal and multi‐scale DEM matching . The existing DEM matching technologies usually fail when DEM is matched without spatial reference because of the insufficient feature points or the vast differences of resolution ,terrain features and some others .Therefore ,the method ,fuzzy matching of DEM without spatial reference based on rectangle features ,is proposed in this paper .First ,five images similarity indicators were constructed based on rectangle features and then the statistical analysis was used respectively in every indicator ’ s classification performance .Second ,the classifier ,to discriminate whether matched or not ,was obtained after using Gentle AdaBoost algorithm training massive samples .Finally ,a model of DEM image fuzzy matching was established .The method used in this paper shows a capability in solving two difficult problems ,the lack of feature points when DEM is matched without spatial reference and low similarity between DEM data ,by verifying of four typical data .All these above may provide precise DEM matching a fine initial condition .%多源、多时相与多尺度 DEM 的匹配是 DEM （Digital Elevation Model）应用中的关键技术。对无空间参考的 DEM 进行匹配时，若特征点不足或匹配 DEM 之间的分辨率、地形特征等差异较大，已有的 DEM 匹配技术往往难以成功。为此，这里提出了基于矩形特征的模糊匹配方法。首先利用矩形特征构建了五种图像相似度指标 BRFSI 并统计分析了各个指标的分类性能；然后，使用 Gentle AdaBoost 算法对大量样本进行训练后，得到了用于判定“匹配”与“非匹配”的分类器；最后建立了进行 DEM 图像模糊匹配的匹配模型。经过四种典型数据验证，该方法很好地解决了无空间参考 DEM 匹配中特征点不足、DEM 之间相似度较低的难点问题，为进一步的 DEM 精匹配提供了良好的初始条件。